1 // Copyright 2014 The Chromium Authors. All rights reserved.
2 // Use of this source code is governed by a BSD-style license that can be
3 // found in the LICENSE file.
4
5 #include <stdio.h>
6 #include <stdlib.h>
7 #include <algorithm> // for min()
8
9 #include "base/atomicops.h"
10 #include "testing/gtest/include/gtest/gtest.h"
11
12 // Number of bits in a size_t.
13 static const int kSizeBits = 8 * sizeof(size_t);
14 // The maximum size of a size_t.
15 static const size_t kMaxSize = ~static_cast<size_t>(0);
16 // Maximum positive size of a size_t if it were signed.
17 static const size_t kMaxSignedSize = ((size_t(1) << (kSizeBits-1)) - 1);
18 // An allocation size which is not too big to be reasonable.
19 static const size_t kNotTooBig = 100000;
20 // An allocation size which is just too big.
21 static const size_t kTooBig = ~static_cast<size_t>(0);
22
23 namespace {
24
25 using std::min;
26
27 // Fill a buffer of the specified size with a predetermined pattern
Fill(unsigned char * buffer,int n)28 static void Fill(unsigned char* buffer, int n) {
29 for (int i = 0; i < n; i++) {
30 buffer[i] = (i & 0xff);
31 }
32 }
33
34 // Check that the specified buffer has the predetermined pattern
35 // generated by Fill()
Valid(unsigned char * buffer,int n)36 static bool Valid(unsigned char* buffer, int n) {
37 for (int i = 0; i < n; i++) {
38 if (buffer[i] != (i & 0xff)) {
39 return false;
40 }
41 }
42 return true;
43 }
44
45 // Check that a buffer is completely zeroed.
IsZeroed(unsigned char * buffer,int n)46 static bool IsZeroed(unsigned char* buffer, int n) {
47 for (int i = 0; i < n; i++) {
48 if (buffer[i] != 0) {
49 return false;
50 }
51 }
52 return true;
53 }
54
55 // Check alignment
CheckAlignment(void * p,int align)56 static void CheckAlignment(void* p, int align) {
57 EXPECT_EQ(0, reinterpret_cast<uintptr_t>(p) & (align-1));
58 }
59
60 // Return the next interesting size/delta to check. Returns -1 if no more.
NextSize(int size)61 static int NextSize(int size) {
62 if (size < 100)
63 return size+1;
64
65 if (size < 100000) {
66 // Find next power of two
67 int power = 1;
68 while (power < size)
69 power <<= 1;
70
71 // Yield (power-1, power, power+1)
72 if (size < power-1)
73 return power-1;
74
75 if (size == power-1)
76 return power;
77
78 assert(size == power);
79 return power+1;
80 } else {
81 return -1;
82 }
83 }
84
85 template <class AtomicType>
TestAtomicIncrement()86 static void TestAtomicIncrement() {
87 // For now, we just test single threaded execution
88
89 // use a guard value to make sure the NoBarrier_AtomicIncrement doesn't go
90 // outside the expected address bounds. This is in particular to
91 // test that some future change to the asm code doesn't cause the
92 // 32-bit NoBarrier_AtomicIncrement to do the wrong thing on 64-bit machines.
93 struct {
94 AtomicType prev_word;
95 AtomicType count;
96 AtomicType next_word;
97 } s;
98
99 AtomicType prev_word_value, next_word_value;
100 memset(&prev_word_value, 0xFF, sizeof(AtomicType));
101 memset(&next_word_value, 0xEE, sizeof(AtomicType));
102
103 s.prev_word = prev_word_value;
104 s.count = 0;
105 s.next_word = next_word_value;
106
107 EXPECT_EQ(base::subtle::NoBarrier_AtomicIncrement(&s.count, 1), 1);
108 EXPECT_EQ(s.count, 1);
109 EXPECT_EQ(s.prev_word, prev_word_value);
110 EXPECT_EQ(s.next_word, next_word_value);
111
112 EXPECT_EQ(base::subtle::NoBarrier_AtomicIncrement(&s.count, 2), 3);
113 EXPECT_EQ(s.count, 3);
114 EXPECT_EQ(s.prev_word, prev_word_value);
115 EXPECT_EQ(s.next_word, next_word_value);
116
117 EXPECT_EQ(base::subtle::NoBarrier_AtomicIncrement(&s.count, 3), 6);
118 EXPECT_EQ(s.count, 6);
119 EXPECT_EQ(s.prev_word, prev_word_value);
120 EXPECT_EQ(s.next_word, next_word_value);
121
122 EXPECT_EQ(base::subtle::NoBarrier_AtomicIncrement(&s.count, -3), 3);
123 EXPECT_EQ(s.count, 3);
124 EXPECT_EQ(s.prev_word, prev_word_value);
125 EXPECT_EQ(s.next_word, next_word_value);
126
127 EXPECT_EQ(base::subtle::NoBarrier_AtomicIncrement(&s.count, -2), 1);
128 EXPECT_EQ(s.count, 1);
129 EXPECT_EQ(s.prev_word, prev_word_value);
130 EXPECT_EQ(s.next_word, next_word_value);
131
132 EXPECT_EQ(base::subtle::NoBarrier_AtomicIncrement(&s.count, -1), 0);
133 EXPECT_EQ(s.count, 0);
134 EXPECT_EQ(s.prev_word, prev_word_value);
135 EXPECT_EQ(s.next_word, next_word_value);
136
137 EXPECT_EQ(base::subtle::NoBarrier_AtomicIncrement(&s.count, -1), -1);
138 EXPECT_EQ(s.count, -1);
139 EXPECT_EQ(s.prev_word, prev_word_value);
140 EXPECT_EQ(s.next_word, next_word_value);
141
142 EXPECT_EQ(base::subtle::NoBarrier_AtomicIncrement(&s.count, -4), -5);
143 EXPECT_EQ(s.count, -5);
144 EXPECT_EQ(s.prev_word, prev_word_value);
145 EXPECT_EQ(s.next_word, next_word_value);
146
147 EXPECT_EQ(base::subtle::NoBarrier_AtomicIncrement(&s.count, 5), 0);
148 EXPECT_EQ(s.count, 0);
149 EXPECT_EQ(s.prev_word, prev_word_value);
150 EXPECT_EQ(s.next_word, next_word_value);
151 }
152
153
154 #define NUM_BITS(T) (sizeof(T) * 8)
155
156
157 template <class AtomicType>
TestCompareAndSwap()158 static void TestCompareAndSwap() {
159 AtomicType value = 0;
160 AtomicType prev = base::subtle::NoBarrier_CompareAndSwap(&value, 0, 1);
161 EXPECT_EQ(1, value);
162 EXPECT_EQ(0, prev);
163
164 // Use test value that has non-zero bits in both halves, more for testing
165 // 64-bit implementation on 32-bit platforms.
166 const AtomicType k_test_val = (static_cast<uint64_t>(1) <<
167 (NUM_BITS(AtomicType) - 2)) + 11;
168 value = k_test_val;
169 prev = base::subtle::NoBarrier_CompareAndSwap(&value, 0, 5);
170 EXPECT_EQ(k_test_val, value);
171 EXPECT_EQ(k_test_val, prev);
172
173 value = k_test_val;
174 prev = base::subtle::NoBarrier_CompareAndSwap(&value, k_test_val, 5);
175 EXPECT_EQ(5, value);
176 EXPECT_EQ(k_test_val, prev);
177 }
178
179
180 template <class AtomicType>
TestAtomicExchange()181 static void TestAtomicExchange() {
182 AtomicType value = 0;
183 AtomicType new_value = base::subtle::NoBarrier_AtomicExchange(&value, 1);
184 EXPECT_EQ(1, value);
185 EXPECT_EQ(0, new_value);
186
187 // Use test value that has non-zero bits in both halves, more for testing
188 // 64-bit implementation on 32-bit platforms.
189 const AtomicType k_test_val = (static_cast<uint64_t>(1) <<
190 (NUM_BITS(AtomicType) - 2)) + 11;
191 value = k_test_val;
192 new_value = base::subtle::NoBarrier_AtomicExchange(&value, k_test_val);
193 EXPECT_EQ(k_test_val, value);
194 EXPECT_EQ(k_test_val, new_value);
195
196 value = k_test_val;
197 new_value = base::subtle::NoBarrier_AtomicExchange(&value, 5);
198 EXPECT_EQ(5, value);
199 EXPECT_EQ(k_test_val, new_value);
200 }
201
202
203 template <class AtomicType>
TestAtomicIncrementBounds()204 static void TestAtomicIncrementBounds() {
205 // Test increment at the half-width boundary of the atomic type.
206 // It is primarily for testing at the 32-bit boundary for 64-bit atomic type.
207 AtomicType test_val = static_cast<uint64_t>(1) << (NUM_BITS(AtomicType) / 2);
208 AtomicType value = test_val - 1;
209 AtomicType new_value = base::subtle::NoBarrier_AtomicIncrement(&value, 1);
210 EXPECT_EQ(test_val, value);
211 EXPECT_EQ(value, new_value);
212
213 base::subtle::NoBarrier_AtomicIncrement(&value, -1);
214 EXPECT_EQ(test_val - 1, value);
215 }
216
217 // This is a simple sanity check that values are correct. Not testing
218 // atomicity
219 template <class AtomicType>
TestStore()220 static void TestStore() {
221 const AtomicType kVal1 = static_cast<AtomicType>(0xa5a5a5a5a5a5a5a5LL);
222 const AtomicType kVal2 = static_cast<AtomicType>(-1);
223
224 AtomicType value;
225
226 base::subtle::NoBarrier_Store(&value, kVal1);
227 EXPECT_EQ(kVal1, value);
228 base::subtle::NoBarrier_Store(&value, kVal2);
229 EXPECT_EQ(kVal2, value);
230
231 base::subtle::Acquire_Store(&value, kVal1);
232 EXPECT_EQ(kVal1, value);
233 base::subtle::Acquire_Store(&value, kVal2);
234 EXPECT_EQ(kVal2, value);
235
236 base::subtle::Release_Store(&value, kVal1);
237 EXPECT_EQ(kVal1, value);
238 base::subtle::Release_Store(&value, kVal2);
239 EXPECT_EQ(kVal2, value);
240 }
241
242 // This is a simple sanity check that values are correct. Not testing
243 // atomicity
244 template <class AtomicType>
TestLoad()245 static void TestLoad() {
246 const AtomicType kVal1 = static_cast<AtomicType>(0xa5a5a5a5a5a5a5a5LL);
247 const AtomicType kVal2 = static_cast<AtomicType>(-1);
248
249 AtomicType value;
250
251 value = kVal1;
252 EXPECT_EQ(kVal1, base::subtle::NoBarrier_Load(&value));
253 value = kVal2;
254 EXPECT_EQ(kVal2, base::subtle::NoBarrier_Load(&value));
255
256 value = kVal1;
257 EXPECT_EQ(kVal1, base::subtle::Acquire_Load(&value));
258 value = kVal2;
259 EXPECT_EQ(kVal2, base::subtle::Acquire_Load(&value));
260
261 value = kVal1;
262 EXPECT_EQ(kVal1, base::subtle::Release_Load(&value));
263 value = kVal2;
264 EXPECT_EQ(kVal2, base::subtle::Release_Load(&value));
265 }
266
267 template <class AtomicType>
TestAtomicOps()268 static void TestAtomicOps() {
269 TestCompareAndSwap<AtomicType>();
270 TestAtomicExchange<AtomicType>();
271 TestAtomicIncrementBounds<AtomicType>();
272 TestStore<AtomicType>();
273 TestLoad<AtomicType>();
274 }
275
TestCalloc(size_t n,size_t s,bool ok)276 static void TestCalloc(size_t n, size_t s, bool ok) {
277 char* p = reinterpret_cast<char*>(calloc(n, s));
278 if (!ok) {
279 EXPECT_EQ(NULL, p) << "calloc(n, s) should not succeed";
280 } else {
281 EXPECT_NE(reinterpret_cast<void*>(NULL), p) <<
282 "calloc(n, s) should succeed";
283 for (int i = 0; i < n*s; i++) {
284 EXPECT_EQ('\0', p[i]);
285 }
286 free(p);
287 }
288 }
289
290
291 // A global test counter for number of times the NewHandler is called.
292 static int news_handled = 0;
TestNewHandler()293 static void TestNewHandler() {
294 ++news_handled;
295 throw std::bad_alloc();
296 }
297
298 // Because we compile without exceptions, we expect these will not throw.
TestOneNewWithoutExceptions(void * (* func)(size_t),bool should_throw)299 static void TestOneNewWithoutExceptions(void* (*func)(size_t),
300 bool should_throw) {
301 // success test
302 try {
303 void* ptr = (*func)(kNotTooBig);
304 EXPECT_NE(reinterpret_cast<void*>(NULL), ptr) <<
305 "allocation should not have failed.";
306 } catch(...) {
307 EXPECT_EQ(0, 1) << "allocation threw unexpected exception.";
308 }
309
310 // failure test
311 try {
312 void* rv = (*func)(kTooBig);
313 EXPECT_EQ(NULL, rv);
314 EXPECT_FALSE(should_throw) << "allocation should have thrown.";
315 } catch(...) {
316 EXPECT_TRUE(should_throw) << "allocation threw unexpected exception.";
317 }
318 }
319
TestNothrowNew(void * (* func)(size_t))320 static void TestNothrowNew(void* (*func)(size_t)) {
321 news_handled = 0;
322
323 // test without new_handler:
324 std::new_handler saved_handler = std::set_new_handler(0);
325 TestOneNewWithoutExceptions(func, false);
326
327 // test with new_handler:
328 std::set_new_handler(TestNewHandler);
329 TestOneNewWithoutExceptions(func, true);
330 EXPECT_EQ(news_handled, 1) << "nothrow new_handler was not called.";
331 std::set_new_handler(saved_handler);
332 }
333
334 } // namespace
335
336 //-----------------------------------------------------------------------------
337
TEST(Atomics,AtomicIncrementWord)338 TEST(Atomics, AtomicIncrementWord) {
339 TestAtomicIncrement<AtomicWord>();
340 }
341
TEST(Atomics,AtomicIncrement32)342 TEST(Atomics, AtomicIncrement32) {
343 TestAtomicIncrement<Atomic32>();
344 }
345
TEST(Atomics,AtomicOpsWord)346 TEST(Atomics, AtomicOpsWord) {
347 TestAtomicIncrement<AtomicWord>();
348 }
349
TEST(Atomics,AtomicOps32)350 TEST(Atomics, AtomicOps32) {
351 TestAtomicIncrement<Atomic32>();
352 }
353
TEST(Allocators,Malloc)354 TEST(Allocators, Malloc) {
355 // Try allocating data with a bunch of alignments and sizes
356 for (int size = 1; size < 1048576; size *= 2) {
357 unsigned char* ptr = reinterpret_cast<unsigned char*>(malloc(size));
358 CheckAlignment(ptr, 2); // Should be 2 byte aligned
359 Fill(ptr, size);
360 EXPECT_TRUE(Valid(ptr, size));
361 free(ptr);
362 }
363 }
364
TEST(Allocators,Calloc)365 TEST(Allocators, Calloc) {
366 TestCalloc(0, 0, true);
367 TestCalloc(0, 1, true);
368 TestCalloc(1, 1, true);
369 TestCalloc(1<<10, 0, true);
370 TestCalloc(1<<20, 0, true);
371 TestCalloc(0, 1<<10, true);
372 TestCalloc(0, 1<<20, true);
373 TestCalloc(1<<20, 2, true);
374 TestCalloc(2, 1<<20, true);
375 TestCalloc(1000, 1000, true);
376
377 TestCalloc(kMaxSize, 2, false);
378 TestCalloc(2, kMaxSize, false);
379 TestCalloc(kMaxSize, kMaxSize, false);
380
381 TestCalloc(kMaxSignedSize, 3, false);
382 TestCalloc(3, kMaxSignedSize, false);
383 TestCalloc(kMaxSignedSize, kMaxSignedSize, false);
384 }
385
TEST(Allocators,New)386 TEST(Allocators, New) {
387 TestNothrowNew(&::operator new);
388 TestNothrowNew(&::operator new[]);
389 }
390
391 // This makes sure that reallocing a small number of bytes in either
392 // direction doesn't cause us to allocate new memory.
TEST(Allocators,Realloc1)393 TEST(Allocators, Realloc1) {
394 int start_sizes[] = { 100, 1000, 10000, 100000 };
395 int deltas[] = { 1, -2, 4, -8, 16, -32, 64, -128 };
396
397 for (int s = 0; s < sizeof(start_sizes)/sizeof(*start_sizes); ++s) {
398 void* p = malloc(start_sizes[s]);
399 ASSERT_TRUE(p);
400 // The larger the start-size, the larger the non-reallocing delta.
401 for (int d = 0; d < s*2; ++d) {
402 void* new_p = realloc(p, start_sizes[s] + deltas[d]);
403 ASSERT_EQ(p, new_p); // realloc should not allocate new memory
404 }
405 // Test again, but this time reallocing smaller first.
406 for (int d = 0; d < s*2; ++d) {
407 void* new_p = realloc(p, start_sizes[s] - deltas[d]);
408 ASSERT_EQ(p, new_p); // realloc should not allocate new memory
409 }
410 free(p);
411 }
412 }
413
TEST(Allocators,Realloc2)414 TEST(Allocators, Realloc2) {
415 for (int src_size = 0; src_size >= 0; src_size = NextSize(src_size)) {
416 for (int dst_size = 0; dst_size >= 0; dst_size = NextSize(dst_size)) {
417 unsigned char* src = reinterpret_cast<unsigned char*>(malloc(src_size));
418 Fill(src, src_size);
419 unsigned char* dst =
420 reinterpret_cast<unsigned char*>(realloc(src, dst_size));
421 EXPECT_TRUE(Valid(dst, min(src_size, dst_size)));
422 Fill(dst, dst_size);
423 EXPECT_TRUE(Valid(dst, dst_size));
424 if (dst != NULL) free(dst);
425 }
426 }
427
428 // Now make sure realloc works correctly even when we overflow the
429 // packed cache, so some entries are evicted from the cache.
430 // The cache has 2^12 entries, keyed by page number.
431 const int kNumEntries = 1 << 14;
432 int** p = reinterpret_cast<int**>(malloc(sizeof(*p) * kNumEntries));
433 int sum = 0;
434 for (int i = 0; i < kNumEntries; i++) {
435 // no page size is likely to be bigger than 8192?
436 p[i] = reinterpret_cast<int*>(malloc(8192));
437 p[i][1000] = i; // use memory deep in the heart of p
438 }
439 for (int i = 0; i < kNumEntries; i++) {
440 p[i] = reinterpret_cast<int*>(realloc(p[i], 9000));
441 }
442 for (int i = 0; i < kNumEntries; i++) {
443 sum += p[i][1000];
444 free(p[i]);
445 }
446 EXPECT_EQ(kNumEntries/2 * (kNumEntries - 1), sum); // assume kNE is even
447 free(p);
448 }
449
TEST(Allocators,ReallocZero)450 TEST(Allocators, ReallocZero) {
451 // Test that realloc to zero does not return NULL.
452 for (int size = 0; size >= 0; size = NextSize(size)) {
453 char* ptr = reinterpret_cast<char*>(malloc(size));
454 EXPECT_NE(static_cast<char*>(NULL), ptr);
455 ptr = reinterpret_cast<char*>(realloc(ptr, 0));
456 EXPECT_NE(static_cast<char*>(NULL), ptr);
457 if (ptr)
458 free(ptr);
459 }
460 }
461
462 #ifdef WIN32
463 // Test recalloc
TEST(Allocators,Recalloc)464 TEST(Allocators, Recalloc) {
465 for (int src_size = 0; src_size >= 0; src_size = NextSize(src_size)) {
466 for (int dst_size = 0; dst_size >= 0; dst_size = NextSize(dst_size)) {
467 unsigned char* src =
468 reinterpret_cast<unsigned char*>(_recalloc(NULL, 1, src_size));
469 EXPECT_TRUE(IsZeroed(src, src_size));
470 Fill(src, src_size);
471 unsigned char* dst =
472 reinterpret_cast<unsigned char*>(_recalloc(src, 1, dst_size));
473 EXPECT_TRUE(Valid(dst, min(src_size, dst_size)));
474 Fill(dst, dst_size);
475 EXPECT_TRUE(Valid(dst, dst_size));
476 if (dst != NULL)
477 free(dst);
478 }
479 }
480 }
481
482 // Test windows specific _aligned_malloc() and _aligned_free() methods.
TEST(Allocators,AlignedMalloc)483 TEST(Allocators, AlignedMalloc) {
484 // Try allocating data with a bunch of alignments and sizes
485 static const int kTestAlignments[] = {8, 16, 256, 4096, 8192, 16384};
486 for (int size = 1; size > 0; size = NextSize(size)) {
487 for (int i = 0; i < ARRAYSIZE(kTestAlignments); ++i) {
488 unsigned char* ptr = static_cast<unsigned char*>(
489 _aligned_malloc(size, kTestAlignments[i]));
490 CheckAlignment(ptr, kTestAlignments[i]);
491 Fill(ptr, size);
492 EXPECT_TRUE(Valid(ptr, size));
493
494 // Make a second allocation of the same size and alignment to prevent
495 // allocators from passing this test by accident. Per jar, tcmalloc
496 // provides allocations for new (never before seen) sizes out of a thread
497 // local heap of a given "size class." Each time the test requests a new
498 // size, it will usually get the first element of a span, which is a
499 // 4K aligned allocation.
500 unsigned char* ptr2 = static_cast<unsigned char*>(
501 _aligned_malloc(size, kTestAlignments[i]));
502 CheckAlignment(ptr2, kTestAlignments[i]);
503 Fill(ptr2, size);
504 EXPECT_TRUE(Valid(ptr2, size));
505
506 // Should never happen, but sanity check just in case.
507 ASSERT_NE(ptr, ptr2);
508 _aligned_free(ptr);
509 _aligned_free(ptr2);
510 }
511 }
512 }
513
514 #endif
515
516
main(int argc,char ** argv)517 int main(int argc, char** argv) {
518 testing::InitGoogleTest(&argc, argv);
519 return RUN_ALL_TESTS();
520 }
521